Self-Driving Car Engineer Nanodegree¶
Deep Learning¶
Project: Build a Traffic Sign Recognition Classifier¶
In this notebook, a template is provided for you to implement your functionality in stages, which is required to successfully complete this project. If additional code is required that cannot be included in the notebook, be sure that the Python code is successfully imported and included in your submission if necessary.
Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.
In addition to implementing code, there is a writeup to complete. The writeup should be completed in a separate file, which can be either a markdown file or a pdf document. There is a write up template that can be used to guide the writing process. Completing the code template and writeup template will cover all of the rubric points for this project.
The rubric contains "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. The stand out suggestions are optional. If you decide to pursue the "stand out suggestions", you can include the code in this Ipython notebook and also discuss the results in the writeup file.
Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. In addition, Markdown cells can be edited by typically double-clicking the cell to enter edit mode.
Step 0: Load The Data¶
In this module, we need to initialize and import the train_data, valid_data and test_data Using panda is easily to access the csv file.
# Load pickled data
import cv2
import pickle
import numpy as np
from matplotlib import pyplot
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
from pandas.io.parsers import read_csv
from skimage import transform as transf
# TODO: Fill this in based on where you saved the training and testing data
training_file = 'dataset/train.p'
validation_file='dataset/valid.p'
testing_file = 'dataset/test.p'
with open(training_file, mode='rb') as f:
train = pickle.load(f)
with open(validation_file, mode='rb') as f:
valid = pickle.load(f)
with open(testing_file, mode='rb') as f:
test = pickle.load(f)
sign_names = read_csv("signnames.csv").values[:, 1]
X_train, y_train = train['features'], train['labels']
X_valid, y_valid = valid['features'], valid['labels']
X_test, y_test = test['features'], test['labels']
Step 1: Dataset Summary & Exploration¶
The pickled data is a dictionary with 4 key/value pairs:
'features'is a 4D array containing raw pixel data of the traffic sign images, (num examples, width, height, channels).'labels'is a 1D array containing the label/class id of the traffic sign. The filesignnames.csvcontains id -> name mappings for each id.'sizes'is a list containing tuples, (width, height) representing the original width and height the image.'coords'is a list containing tuples, (x1, y1, x2, y2) representing coordinates of a bounding box around the sign in the image. THESE COORDINATES ASSUME THE ORIGINAL IMAGE. THE PICKLED DATA CONTAINS RESIZED VERSIONS (32 by 32) OF THESE IMAGES
Complete the basic data summary below. Use python, numpy and/or pandas methods to calculate the data summary rather than hard coding the results. For example, the pandas shape method might be useful for calculating some of the summary results.
Provide a Basic Summary of the Data Set Using Python, Numpy and/or Pandas¶
In this modul, we will get the shape of train_data, valid_data and test_data .
### Replace each question mark with the appropriate value.
### Use python, pandas or numpy methods rather than hard coding the results
# TODO: Number of training examples
n_train = X_train.shape[0]
# TODO: Number of validation examples
n_validation = X_valid.shape[0]
# TODO: Number of testing examples.
n_test = X_test.shape[0]
# TODO: What's the shape of an traffic sign image?
image_shape = X_train[0].shape
# TODO: How many unique classes/labels there are in the dataset.
# By using the np.unique we can get the numbers , indices and counts of different classes in original data.
classes, class_indices, class_counts = np.unique(y_train, return_index=True, return_counts=True)
n_classes = len(classes)
print("Number of training examples =", n_train)
print("Number of testing examples =", n_test)
print("Image data shape =", image_shape)
print("Number of classes =", n_classes)
# By using plt we can draw the histogram of class distribution of data
plt.figure(0, figsize = (10,5))
train_classes, train_counts = np.unique(y_train, return_counts = True)
plt.bar(train_classes, train_counts)
plt.title('Training Set Class Distribution', fontsize=22)
plt.xlabel('Class Number', fontsize=20)
plt.ylabel('Number of Occurances', fontsize=20)
plt.tick_params(labelsize=16)
plt.grid(linestyle=':')
Include an exploratory visualization of the dataset¶
Visualize the German Traffic Signs Dataset using the pickled file(s). This is open ended, suggestions include: plotting traffic sign images, plotting the count of each sign, etc.
The Matplotlib examples and gallery pages are a great resource for doing visualizations in Python.
NOTE: It's recommended you start with something simple first. If you wish to do more, come back to it after you've completed the rest of the sections. It can be interesting to look at the distribution of classes in the training, validation and test set. Is the distribution the same? Are there more examples of some classes than others?
### Data exploration visualization code goes here.
### Feel free to use as many code cells as needed.
import matplotlib.pyplot as plt
# Visualizations will be shown in the notebook.
%matplotlib inline
for each_c, each_c_i, each_c_c in zip(classes, class_indices, class_counts):
print(each_c, ". Class : ", sign_names[each_c] )
figs,axes=plt.subplots(1,10,figsize=(10,1))
figs.subplots_adjust(left = 0, right = 1, bottom = 0, top = 1, hspace = 0.05, wspace = 0.05)
for i in range(10):
random_indices = np.random.randint(each_c_i, each_c_i+each_c_c, 10)
axes[i].imshow(X_train[random_indices[i],:,:,:])
axes[i].axis('off')
pyplot.show()
Step 2: Design and Test a Model Architecture¶
Design and implement a deep learning model that learns to recognize traffic signs. Train and test your model on the German Traffic Sign Dataset.
The LeNet-5 implementation shown in the classroom at the end of the CNN lesson is a solid starting point. You'll have to change the number of classes and possibly the preprocessing, but aside from that it's plug and play!
With the LeNet-5 solution from the lecture, you should expect a validation set accuracy of about 0.89. To meet specifications, the validation set accuracy will need to be at least 0.93. It is possible to get an even higher accuracy, but 0.93 is the minimum for a successful project submission.
There are various aspects to consider when thinking about this problem:
- Neural network architecture (is the network over or underfitting?)
- Play around preprocessing techniques (normalization, rgb to grayscale, etc)
- Number of examples per label (some have more than others).
- Generate fake data.
Here is an example of a published baseline model on this problem. It's not required to be familiar with the approach used in the paper but, it's good practice to try to read papers like these.
Pre-process the Data Set (normalization, grayscale, etc.)¶
Minimally, the image data should be normalized so that the data has mean zero and equal variance. For image data, (pixel - 128)/ 128 is a quick way to approximately normalize the data and can be used in this project.
Other pre-processing steps are optional. You can try different techniques to see if it improves performance.
Use the code cell (or multiple code cells, if necessary) to implement the first step of your project.
In this modul, we will perform data augmentation. It include rotation, translation, scaling, shear and light-change operation.
### Preprocess the data here. It is required to normalize the data. Other preprocessing steps could include
### converting to grayscale, etc.
### Feel free to use as many code cells as needed.
def random_transform(img,angle_range=[-10,10],
scale_range=[0.8,1.2],
translation_range=[-5,5],
shear_range=[-0.2,0.2]):
# Generate random parameter values
rotated_img = 255*transf.rotate(img, angle=np.random.uniform(angle_range[0], angle_range[1]), mode='edge')
translated_img = transf.warp(rotated_img, transf.SimilarityTransform(translation=(np.random.uniform(translation_range[0], translation_range[1]), np.random.uniform(translation_range[0], translation_range[1]))),mode='edge')
scaled_img = transf.warp(translated_img,
transf.AffineTransform(scale=(np.random.uniform(scale_range[0], scale_range[1]), np.random.uniform(scale_range[0],scale_range[1])), shear=np.random.uniform(shear_range[0],shear_range[1])),
mode='edge')
return scaled_img.astype(np.uint8)
def random_brightness(image,min_bright=0.25,max_bright=1):
image = image.astype(np.uint8)
image1 = cv2.cvtColor(image,cv2.COLOR_RGB2HSV)
random_bright = np.random.uniform(min_bright, max_bright)
image1[:,:,2] = image1[:,:,2]*random_bright
image1 = cv2.cvtColor(image1,cv2.COLOR_HSV2RGB)
return image1
# Combine all data augmentation methods.
def transform_image(image):
result=random_transform(random_brightness(image)).astype(np.uint8)
return result
# visualization of results of data augmentation
for each_c, each_c_i, each_c_c in zip(classes, class_indices, class_counts):
print(each_c, ". Class : ", sign_names[each_c] )
figs,axes=plt.subplots(1,10,figsize=(10,1))
figs.subplots_adjust(left = 0, right = 1, bottom = 0, top = 1, hspace = 0.05, wspace = 0.05)
for i in range(10):
random_indices = np.random.randint(each_c_i, each_c_i+each_c_c, 10)
axes[i].imshow(transform_image(X_train[random_indices[i],:,:,:]))
axes[i].axis('off')
pyplot.show()
# Perform data augmentation
# First: random transform images
# Seconde: generate more transformed images to make the distribution of classes same.
def data_augmentation(X_train, y_train, total_per_class =5000):
classes, class_indices, class_counts = np.unique(y_train, return_index=True, return_counts=True)
height, width, channels = X_train[0].shape
X_total = np.empty([0, X_train.shape[1], X_train.shape[2], X_train.shape[3]], dtype = np.float32)
y_total = np.empty([0], dtype = y_train.dtype)
for each_c,each_c_i, each_c_c in zip(classes,class_indices, class_counts):
num_extra=total_per_class-each_c_c
##Copy over the current data for the given class
X_orig = X_train[y_train == each_c]
y_orig = y_train[y_train == each_c]
##Add original data to the new dataset
X_total = np.append(X_total, X_orig, axis=0)
y_total=np.append(y_total,y_orig)
print(each_c," : ",each_c_c," + ", num_extra)
X_extra = np.empty([num_extra, X_train.shape[1], X_train.shape[2], X_train.shape[3]], dtype = np.float32)
y_extra=np.empty([num_extra],dtype = y_train.dtype)
for i in range(num_extra):
rand_id=np.random.randint(each_c_i,each_c_i+each_c_c)
X_extra[i,:,:,:] = transform_image(X_train[rand_id]).reshape((1, height, width, channels))
y_extra[i]=each_c
X_total = np.append(X_total,X_extra, axis=0)
y_total= np.append(y_total,y_extra)
return X_total.astype(np.uint8), y_total
# Save the augmentation data
# I just generate max 4000 images, because large samples will slow the computation speed of my computer. You can set more bigger.
# Dont need run it every time, just one time
X_train_aug,y_train_aug=data_augmentation(X_train, y_train, total_per_class =4000)
np.savez_compressed('dataset/AugTrafficdata', X_train_aug = X_train_aug, y_train_aug = y_train_aug)
# visualization of new data after data augmentation
plt.figure(0, figsize = (10,5))
total_classes, total_indices,total_counts = np.unique(y_train_aug, return_index=True,return_counts = True)
plt.bar(total_classes, total_counts)
plt.title('Augmented Set Class Distribution', fontsize=22)
plt.xlabel('Class Number', fontsize=20)
plt.ylabel('Number of Occurances', fontsize=20)
plt.tick_params(labelsize=16)
plt.grid(linestyle=':')
for each_c, each_c_i, each_c_c in zip(total_classes, total_indices, total_counts):
print(each_c, ". Class : ", sign_names[each_c] )
figs,axes=plt.subplots(1,10,figsize=(10,1))
figs.subplots_adjust(left = 0, right = 1, bottom = 0, top = 1, hspace = 0.05, wspace = 0.05)
for i in range(10):
random_indices = np.random.randint(each_c_i, each_c_i+each_c_c, 10)
axes[i].imshow(X_train_aug[random_indices[i],:,:,:])
axes[i].axis('off')
pyplot.show()
# From here we start to prepare and train deep learning model
from sklearn.utils import shuffle
import numpy as np
import tensorflow as tf
import os
import cv2
loaded = np.load('dataset/AugTrafficdata.npz')
X_train_aug = loaded['X_train_aug']
y_train_aug = loaded['y_train_aug']
print("Augmented Dataset size X")
print(X_train_aug.shape)
print("Augmented Dataset size y")
print(y_train_aug.shape)
plt.figure(0, figsize = (10,5))
valid_classes, valid_counts = np.unique(y_valid, return_counts = True)
plt.bar(valid_classes, valid_counts)
plt.title('Vaild Set Class Distribution', fontsize=22)
plt.xlabel('Class Number', fontsize=20)
plt.ylabel('Number of Occurances', fontsize=20)
plt.tick_params(labelsize=16)
plt.grid(linestyle=':')
plt.figure(0, figsize = (10,5))
test_classes, test_counts = np.unique(y_test, return_counts = True)
plt.bar(test_classes, test_counts)
plt.title('Test Set Class Distribution', fontsize=22)
plt.xlabel('Class Number', fontsize=20)
plt.ylabel('Number of Occurances', fontsize=20)
plt.tick_params(labelsize=16)
plt.grid(linestyle=':')
# Preprocessing the images by using histogram equalization
def equalize_histogram(image):
image = image.astype(np.uint8)
image_yuv = cv2.cvtColor(image, cv2.COLOR_BGR2YUV)
# equalize the histogram of the Y channel
clahe = cv2.createCLAHE(clipLimit=10.0, tileGridSize=(8,8))
image_yuv[:,:,0] = clahe.apply(image_yuv[:,:,0])
# convert the YUV image back to RGB format
image_output = cv2.cvtColor(image_yuv, cv2.COLOR_YUV2BGR)
return image_output
# Perform images normalisation
def normalisation(X):
X_n = np.empty(shape=X.shape)
a = 0.1
b = 0.9
for i, img in enumerate(X):
temp = equalize_histogram(img)
X_max = np.amax(temp)
X_min = np.amin(temp)
X_n[i] = a + (temp - X_min)*(b - a)/(X_max - X_min)
return X_n
# visualization of new data after normalisation
test_image = X_train[np.random.randint(0, X_train.shape[0])]
test_image_eqhist = equalize_histogram(test_image)
test_image_norm = normalisation(np.expand_dims(test_image, axis=0))
f, axarr = plt.subplots(1,3)
axarr[0].imshow(test_image, interpolation='nearest')
axarr[0].set_title("Original image")
axarr[1].imshow(test_image_eqhist, interpolation='nearest')
axarr[1].set_title("+ Histogram" + '\n' + "equalisation")
axarr[2].imshow(test_image_norm[0], interpolation='nearest')
axarr[2].set_title("+ Normalisation")
plt.savefig('Normalisation.jpg', bbox_inches='tight')
plt.show()
# Preprocessing all data by using normalisation and equalization.
X_train = normalisation(X_train)
X_train_aug = normalisation(X_train_aug)
X_valid = normalisation(X_valid)
X_test=normalisation(X_test)
print(X_train.shape)
print(X_valid.shape)
print(X_test.shape)
print(X_train_aug.shape)
# Preprocessing all data by using normalisation and equalization.
aug_classes, aug_indices,aug_counts = np.unique(y_train_aug, return_index=True,return_counts = True)
for each_c, each_c_i, each_c_c in zip(aug_classes, aug_indices, aug_counts):
print(each_c, ". Class : ", sign_names[each_c] )
figs,axes=plt.subplots(1,10,figsize=(10,1))
figs.subplots_adjust(left = 0, right = 1, bottom = 0, top = 1, hspace = 0.05, wspace = 0.05)
for i in range(10):
random_indices = np.random.randint(each_c_i, each_c_i+each_c_c, 10)
axes[i].imshow(X_train_aug[random_indices[i],:,:,:],interpolation='nearest')
axes[i].axis('off')
pyplot.show()
Model Architecture¶
## Train your model here.
## Calculate and report the accuracy on the training and validation set.
## Once a final model architecture is selected,
## the accuracy on the test set should be calculated and reported as well.
## Feel free to use as many code cells as needed.
# here i use densenet model, which is more smaller and efficienter. But it waste too many GPU memory and a little slow.
# this is principle : https://github.com/taki0112/Densenet-Tensorflow
# this is code example of densenet : https://github.com/taki0112/Densenet-Tensorflow/blob/master/MNIST/Densenet_MNIST.py
# i use Mini-batch , MomentumOptimizer ，l2-regulation, weight-decay
def get_weight(shape,name=None):
initial = tf.truncated_normal(shape, stddev=0.01)
return tf.Variable(initial)
def get_bias(shape,name=None):
initial = tf.constant(0.01, shape=shape)
return tf.Variable(initial)
def batch_norm(input, is_train=None):
return tf.contrib.layers.batch_norm(input, scale=True, is_training=is_train, updates_collections=None)
def conv2d(input, in_layers, out_layers, kernel_size, with_bias=False):
W = get_weight([ kernel_size, kernel_size, in_layers, out_layers ])
conv = tf.nn.conv2d(input, W, [ 1, 1, 1, 1 ], padding='SAME')
if with_bias:
return conv + get_bias([ out_layers ])
return conv
def batch_activ_conv(current, in_layers, out_layers, kernel_size, is_training, keep_prob):
current = tf.contrib.layers.batch_norm(current, scale=True, is_training=is_training, updates_collections=None)
current = tf.nn.relu(current)
current = conv2d(current, in_layers, out_layers, kernel_size)
current = tf.nn.dropout(current, keep_prob)
return current
def dense_block(input, num_layers, in_layers, growth, is_training, keep_prob):
output = input
out_layers = in_layers
for index in range(num_layers):
tmp = batch_activ_conv(output, out_layers, growth, 3, is_training, keep_prob)
output = tf.concat( [output, tmp], 3)
out_layers += growth
return output, out_layers
def avg_pool(input, s):
return tf.nn.avg_pool(input, [ 1, s, s, 1 ], [1, s, s, 1 ], 'VALID')
def full_connect(input,input_layers,out_layers):
output = tf.reshape(input, [ -1, input_layers ])
Wfc = get_weight([ input_layers, out_layers ])
bfc = get_bias([ out_layers ])
logits = tf.matmul(output, Wfc) + bfc
return logits
# define densenet model
def DenseNet(X, y , keep_prob, is_train, depth=40):
num_layers = int((depth-4) / 3)
# print('number of features : ',num_layers)
#Inital convolution
conv_1 = conv2d(X, 3, 16, 3)
# print('input shape: ',X.shape)
# print('kern size: 3*3','padding size: 1*1')
# print('conv_1 shape: ',conv_1.shape)
#First DenseBlock
block_1, features = dense_block(conv_1, num_layers, 16, 12, is_train, keep_prob)
bac_1 = batch_activ_conv(block_1, features, features, 1, is_train, keep_prob)
pool_1 = avg_pool(bac_1, 2)
# print('block_1 shape: ',block_1.shape)
# print('features number: ',features)
# print('kern size: 12*12','padding size: 1*1')
# print('bac_1 shape: ',bac_1.shape)
# print('pool_1 shape: ',pool_1.shape)
#Second DenseBlock
block_2, features = dense_block(pool_1, num_layers, features, 12, is_train, keep_prob)
bac_2 = batch_activ_conv(block_2, features, features, 1, is_train, keep_prob)
pool_2 = avg_pool(bac_2, 2)
# print('block_2 shape: ',block_2.shape)
# print('features number: ',features)
# print('kern size: 12*12','padding size: 1*1')
# print('bac_2 shape: ',bac_2.shape)
# print('pool_2 shape: ',pool_2.shape)
#Third DenseBlock
block_3, features = dense_block(pool_2, num_layers, features, 12, is_train, keep_prob)
# print('block_3 shape: ',block_3.shape)
# print('features number: ',features)
bn_3 = batch_norm(block_3,is_train)
relu_3 = tf.nn.relu(bn_3)
pool_3 = avg_pool(relu_3, 8)
final_layers = features
# print('bn_3 shape: ',bn_3.shape)
# print('relu_3 shape: ',relu_3.shape)
# print('pool_3 shape: ',pool_3.shape)
# print('final features: ', final_layers)
logits=full_connect(pool_3,final_layers,43)
# print('output shape',logits.shape)
return logits
Train, Validate and Test the Model¶
A validation set can be used to assess how well the model is performing. A low accuracy on the training and validation sets imply underfitting. A high accuracy on the training set but low accuracy on the validation set implies overfitting.
DenseNet Model Initalising traffic sign classification model
Inital convolution¶
number of features : 12
input shape: (?, 32, 32, 3)
conv kern size: 3x3 padding size: 1x 1
conv_1 shape: (?, 32, 32, 16)
First DenseBlock¶
block_1 shape: (?, 32, 32, 160)
block_1 features number: 160
conv kern size: 12x12 padding size: 1x 1
bac_1 shape: (?, 32, 32, 160)
pool_1 shape: (?, 16, 16, 160) kern size 2*2
Second DenseBlock¶
block_2 shape: (?, 16, 16, 304)
block_2 features number: 304
conv kern size: 12x12 padding size: 1x 1
bac_2 shape: (?, 16, 16, 304)
pool_2 shape: (?, 8, 8, 304)kern size 8x8
Third DenseBlock¶
block_3 shape: (?, 8, 8, 448)
features number: 448
bn_3 shape: (?, 8, 8, 448)
relu_3 shape: (?, 8, 8, 448)
pool_3 shape: (?, 1, 1, 448)kern size 8x8
final features: 448
Output¶
output shape (?, 43)
# Initial parameters
#Dataset parameters
IMAGE_HEIGHT = 32
IMAGE_WIDTH = 32
IMAGE_CHANNELS = 3
NO_CLASSES = 43
# normaly the epochs can be setted to 100, but i think it is too slow, so i just set to 10.
EPOCHS = 30
BATCH_SIZE = 64
LEARNING_RATE =0.01
KEEP_PROB = 0.9
WEIGHT_DECAY=0
# build store folders
def get_dir(directory):
if not os.path.exists(directory):
os.makedirs(directory)
return directory
SAVE_DIR = get_dir('SAVE/')
MODEL_SAVE_DIR = get_dir(SAVE_DIR + 'Models/')
# define all the tensorflow operation in this class, so that we can easily access.
class DenseNetStarter:
def __init__(self, session, learning_rate=0.1, weight_decay=0.0001, keep_prob = 0.8):
self.sess = session
self.learning_rate_init = learning_rate
self.weight_decay = weight_decay
self.input_keep_prob = keep_prob
self.init_graph()
def start_training(self, batch_X, batch_y):
feed_dict = {self.input_image : batch_X, self.y : batch_y, self.keep_prob: self.input_keep_prob ,self.is_train : True, self.learning_rate : self.learning_rate_init}
_, loss, global_step = self.sess.run([self.train_step, self.mean_loss, self.global_step, ], feed_dict=feed_dict)
# show the variation of accuracy and loss each 200 interation
if global_step % 200 ==0:
train_accuracy = self.sess.run(self.accuracy_operation, feed_dict= {self.input_image : batch_X, self.y : batch_y, self.keep_prob :1.0, self.is_train :False})
print("Iteration ", global_step, ": with training loss = ", loss, " and accuracy of " , train_accuracy)
##Code to change the learning rate with the number or epochs needed
def init_graph(self):
print("Initalising traffic sign classification model")
#Define the inputs for the model
with tf.name_scope('Inputs'):
self.input_image = tf.placeholder(tf.float32, shape=[None, IMAGE_HEIGHT, IMAGE_WIDTH, IMAGE_CHANNELS], name='Input_X')
self.y = tf.placeholder(tf.int32, (None), name ='Input_y')
# perform onehot for y-data
self.one_hot_y = tf.one_hot(self.y, NO_CLASSES)
self.is_train = tf.placeholder(tf.bool)
self.keep_prob = tf.placeholder(tf.float32)
self.learning_rate = tf.placeholder(tf.float32)
self.k = tf.placeholder(tf.int32)
##Run batch through model to get logits
self.y_out = DenseNet(self.input_image, self.one_hot_y, self.keep_prob ,self.is_train)
#Calculate the losses
with tf.name_scope('Train'):
self.l2 = tf.add_n([tf.nn.l2_loss(var) for var in tf.trainable_variables()])
self.total_loss = tf.nn.softmax_cross_entropy_with_logits(labels=self.one_hot_y , logits=self.y_out)
self.mean_loss = tf.reduce_mean(self.total_loss)
self.loss = self.mean_loss + self.l2*self.weight_decay
self.correct_prediction = tf.equal(tf.argmax(self.y_out, 1), tf.argmax( self.one_hot_y,1))
self.accuracy_operation = tf.reduce_mean(tf.cast(self.correct_prediction, tf.float32))
# get model's softmax probabilities to show the certainty of the model's predictions
self.probs = tf.nn.softmax(self.y_out)
self.top_probs = tf.nn.top_k(self.probs, self.k)
with tf.name_scope('Optimiser'):
self.global_step = tf.Variable(0, trainable=False, name='global_step')
self.optimizer = tf.train.MomentumOptimizer(self.learning_rate, 0.9, use_nesterov=True)
self.train_step = self.optimizer.minimize(self.loss, global_step = self.global_step)
def evaluate(self, X, y, max_batch = 128):
num_examples = len(X)
total_acc = 0
total_loss =0
for offset in range(0, num_examples, max_batch):
batch_X , batch_y = X[offset:offset + max_batch], y[offset: offset + max_batch]
# keep prob =1.0
feed_dict = {self.input_image : batch_X, self.y : batch_y, self.is_train : False, self.keep_prob: 1.0}
accuracy, loss = self.sess.run([self.accuracy_operation, self.mean_loss ], feed_dict= feed_dict)
total_acc += (accuracy * len(batch_X))
total_loss += (loss * len(batch_X))
return total_acc / num_examples, total_loss / num_examples
def classify_images(self, X, top_k =5):
num_images = len(X)
feed_dict = {self.input_image : X, self.is_train : False, self.keep_prob: 1.0, self.k : top_k}
probs = self.sess.run(self.top_probs, feed_dict= feed_dict)
return probs
# use to store the accuracy and loss of each dataset
valid_hist_acc = []
train_hist_acc = []
valid_hist_loss = []
train_hist_loss = []
tf.reset_default_graph()
with tf.Session() as sess:
print('DenseNet Model')
TrafficNet = DenseNetStarter(sess, LEARNING_RATE, WEIGHT_DECAY, KEEP_PROB)
Saver = tf.train.Saver(max_to_keep=100)
sess.run(tf.global_variables_initializer())
num_examples = len(X_train_aug)
print("Training...")
print()
for i in range(len(valid_hist_acc), EPOCHS):
X_train_aug, y_train_aug = shuffle(X_train_aug, y_train_aug)
for offset in range(0, num_examples, BATCH_SIZE):
end = offset + BATCH_SIZE
batch_x, batch_y = X_train_aug[offset:end], y_train_aug[offset:end]
TrafficNet.start_training(batch_x, batch_y)
validation_accuracy, valid_loss = TrafficNet.evaluate(X_valid, y_valid, max_batch = BATCH_SIZE)
training_accuracy, training_loss = TrafficNet.evaluate(X_train, y_train, max_batch = BATCH_SIZE)
valid_hist_acc.append(validation_accuracy)
train_hist_acc.append(training_accuracy)
valid_hist_loss.append(valid_loss)
train_hist_loss.append(training_loss)
print("EPOCH {} ...".format(i+1))
print("Validation Accuracy = {:.3f}".format(validation_accuracy))
print("Training Accuracy = {:.3f}".format(training_accuracy))
print()
# plot_training(train_hist, valid_hist)
if i % 1 ==0:
print('Saving model ...')
Saver.save(sess, MODEL_SAVE_DIR + 'densenet.ckpt', global_step=i)
# if validation_accuracy>=0.99:
# break
# plot the diagram of variation of accuracy and loss
def plot_training(train_hist_acc_in, valid_hist_acc_in, train_hist_loss_in, valid_hist_loss_in):
curves_figure = pyplot.figure(figsize = (10, 4))
axis = curves_figure.add_subplot(1,2,1)
axis.plot(train_hist_acc_in, label='train')
axis.plot(valid_hist_acc_in, label='valid')
pyplot.grid()
pyplot.legend()
pyplot.xlabel("epoch")
pyplot.ylabel("accuracy")
axis = curves_figure.add_subplot(1,2,2)
axis.plot(train_hist_loss_in, label='train')
axis.plot(valid_hist_loss_in, label='valid')
pyplot.grid()
pyplot.legend()
pyplot.xlabel("epoch")
pyplot.ylabel("cross-entropy")
plot_training( train_hist_acc, valid_hist_acc, train_hist_loss, valid_hist_loss )
# restore the model and test it in test_dataset
# test_data need to be preprocessed
SAVE_DIR = get_dir('SAVE/')
MODEL_SAVE_DIR = get_dir(SAVE_DIR + 'Models/')
MODEL_LOAD_PATH = './' + MODEL_SAVE_DIR + 'densenet.ckpt-29'
tf.reset_default_graph()
with tf.Session() as sess:
TrafficNet = DenseNetStarter(sess, LEARNING_RATE, WEIGHT_DECAY, KEEP_PROB)
Saver = tf.train.Saver(max_to_keep=100)
sess.run(tf.global_variables_initializer())
if MODEL_LOAD_PATH is not None:
Saver.restore(sess, MODEL_LOAD_PATH)
print('Model restored from ' + MODEL_LOAD_PATH)
acc, loss = TrafficNet.evaluate(X_test, y_test, max_batch = BATCH_SIZE)
print("Final DenseNet accuracy on the test set ", acc*100, "%")
tf.reset_default_graph()
Step 3: Test a Model on New Images¶
To give yourself more insight into how your model is working, download at least five pictures of German traffic signs from the web and use your model to predict the traffic sign type.
You may find signnames.csv useful as it contains mappings from the class id (integer) to the actual sign name.
Load and Output the Images¶
### Load the images and plot them here.
### Feel free to use as many code cells as needed.
### Load the images and plot them here.
### Feel free to use as many code cells as needed.
def load_images(folder):
images = []
for filename in os.listdir(folder):
img = cv2.imread(os.path.join(folder,filename))
if img is not None:
images.append(img)
return images
images = load_images('./downloaded_signs')
# Resize images and stack them
for i, image in enumerate(images):
temp = np.copy(image)
temp = cv2.cvtColor(temp, cv2.COLOR_BGR2RGB)
images[i] = np.expand_dims(cv2.resize(temp, (32, 32)), axis=0)
figs,axes=plt.subplots(1,9,figsize=(9,1))
figs.subplots_adjust(left = 0, right = 1, bottom = 0, top = 1, hspace = 0.05, wspace = 0.05)
im_data = np.vstack(images)
for i, image in enumerate(im_data):
axes[i].imshow(image)
axes[i].axis('off')
pyplot.show()
Predict the Sign Type for Each Image¶
### Run the predictions here and use the model to output the prediction for each image.
### Make sure to pre-process the images with the same pre-processing pipeline used earlier.
### Feel free to use as many code cells as needed.
# load sign_names
sign_names = read_csv("signnames.csv").values[:, 1]
# manual label these images
y_data = np.array([7,0,1,5,18,17,25,36,13])
# preprocess these images
im_data_norm = normalisation(im_data)
SAVE_DIR = get_dir('SAVE/')
MODEL_SAVE_DIR = get_dir(SAVE_DIR + 'Models/')
MODEL_LOAD_PATH = './' + MODEL_SAVE_DIR + 'densenet.ckpt-29'
tf.reset_default_graph()
with tf.Session() as sess:
TrafficNet = DenseNetStarter(sess, LEARNING_RATE, WEIGHT_DECAY, KEEP_PROB)
Saver = tf.train.Saver(max_to_keep=100)
sess.run(tf.global_variables_initializer())
if MODEL_LOAD_PATH is not None:
Saver.restore(sess, MODEL_LOAD_PATH)
print('Model restored from ' + MODEL_LOAD_PATH)
values_dense, indices_dense = TrafficNet.classify_images(im_data_norm)
tf.reset_default_graph()
# compare results
for i in range(len(y_data)):
plt.figure()
plt.imshow(im_data_norm[i])
plt.show()
print('Correct label: ' + sign_names[y_data[i]] )
print('Prediced label: ' + sign_names[indices_dense[i]][0] )
print('Match with the Ground Truth : ', str(values_dense[i][0]))
print()
Analyze Performance¶
### Calculate the accuracy for these 5 new images.
### For example, if the model predicted 1 out of 5 signs correctly, it's 20% accurate on these new images.
SAVE_DIR = get_dir('SAVE/')
MODEL_SAVE_DIR = get_dir(SAVE_DIR + 'Models/')
MODEL_LOAD_PATH = './' + MODEL_SAVE_DIR + 'densenet.ckpt-29'
tf.reset_default_graph()
with tf.Session() as sess:
TrafficNet = DenseNetStarter(sess, LEARNING_RATE, WEIGHT_DECAY, KEEP_PROB)
Saver = tf.train.Saver(max_to_keep=100)
sess.run(tf.global_variables_initializer())
if MODEL_LOAD_PATH is not None:
Saver.restore(sess, MODEL_LOAD_PATH)
print('Model restored from ' + MODEL_LOAD_PATH)
acc, loss = TrafficNet.evaluate(im_data_norm, y_data, max_batch = BATCH_SIZE)
print("Final DenseNet accuracy on the test set ", acc*100, "%")
tf.reset_default_graph()
Output Top 5 Softmax Probabilities For Each Image Found on the Web¶
For each of the new images, print out the model's softmax probabilities to show the certainty of the model's predictions (limit the output to the top 5 probabilities for each image). tf.nn.top_k could prove helpful here.
The example below demonstrates how tf.nn.top_k can be used to find the top k predictions for each image.
tf.nn.top_k will return the values and indices (class ids) of the top k predictions. So if k=3, for each sign, it'll return the 3 largest probabilities (out of a possible 43) and the correspoding class ids.
Take this numpy array as an example. The values in the array represent predictions. The array contains softmax probabilities for five candidate images with six possible classes. tf.nn.top_k is used to choose the three classes with the highest probability:
# (5, 6) array
a = np.array([[ 0.24879643, 0.07032244, 0.12641572, 0.34763842, 0.07893497,
0.12789202],
[ 0.28086119, 0.27569815, 0.08594638, 0.0178669 , 0.18063401,
0.15899337],
[ 0.26076848, 0.23664738, 0.08020603, 0.07001922, 0.1134371 ,
0.23892179],
[ 0.11943333, 0.29198961, 0.02605103, 0.26234032, 0.1351348 ,
0.16505091],
[ 0.09561176, 0.34396535, 0.0643941 , 0.16240774, 0.24206137,
0.09155967]])Running it through sess.run(tf.nn.top_k(tf.constant(a), k=3)) produces:
TopKV2(values=array([[ 0.34763842, 0.24879643, 0.12789202],
[ 0.28086119, 0.27569815, 0.18063401],
[ 0.26076848, 0.23892179, 0.23664738],
[ 0.29198961, 0.26234032, 0.16505091],
[ 0.34396535, 0.24206137, 0.16240774]]), indices=array([[3, 0, 5],
[0, 1, 4],
[0, 5, 1],
[1, 3, 5],
[1, 4, 3]], dtype=int32))Looking just at the first row we get [ 0.34763842, 0.24879643, 0.12789202], you can confirm these are the 3 largest probabilities in a. You'll also notice [3, 0, 5] are the corresponding indices.
### Print out the top five softmax probabilities for the predictions on the German traffic sign images found on the web.
### Feel free to use as many code cells as needed.
Note: Once you have completed all of the code implementations and successfully answered each question above, you may finalize your work by exporting the iPython Notebook as an HTML document. You can do this by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.
Step 4 (Optional): Visualize the Neural Network's State with Test Images¶
This Section is not required to complete but acts as an additional excersise for understaning the output of a neural network's weights. While neural networks can be a great learning device they are often referred to as a black box. We can understand what the weights of a neural network look like better by plotting their feature maps. After successfully training your neural network you can see what it's feature maps look like by plotting the output of the network's weight layers in response to a test stimuli image. From these plotted feature maps, it's possible to see what characteristics of an image the network finds interesting. For a sign, maybe the inner network feature maps react with high activation to the sign's boundary outline or to the contrast in the sign's painted symbol.
Provided for you below is the function code that allows you to get the visualization output of any tensorflow weight layer you want. The inputs to the function should be a stimuli image, one used during training or a new one you provided, and then the tensorflow variable name that represents the layer's state during the training process, for instance if you wanted to see what the LeNet lab's feature maps looked like for it's second convolutional layer you could enter conv2 as the tf_activation variable.
For an example of what feature map outputs look like, check out NVIDIA's results in their paper End-to-End Deep Learning for Self-Driving Cars in the section Visualization of internal CNN State. NVIDIA was able to show that their network's inner weights had high activations to road boundary lines by comparing feature maps from an image with a clear path to one without. Try experimenting with a similar test to show that your trained network's weights are looking for interesting features, whether it's looking at differences in feature maps from images with or without a sign, or even what feature maps look like in a trained network vs a completely untrained one on the same sign image.
Your output should look something like this (above)
### Visualize your network's feature maps here.
### Feel free to use as many code cells as needed.
# image_input: the test image being fed into the network to produce the feature maps
# tf_activation: should be a tf variable name used during your training procedure that represents the calculated state of a specific weight layer
# activation_min/max: can be used to view the activation contrast in more detail, by default matplot sets min and max to the actual min and max values of the output
# plt_num: used to plot out multiple different weight feature map sets on the same block, just extend the plt number for each new feature map entry
def outputFeatureMap(image_input, tf_activation, activation_min=-1, activation_max=-1 ,plt_num=1):
# Here make sure to preprocess your image_input in a way your network expects
# with size, normalization, ect if needed
# image_input =
# Note: x should be the same name as your network's tensorflow data placeholder variable
# If you get an error tf_activation is not defined it may be having trouble accessing the variable from inside a function
activation = tf_activation.eval(session=sess,feed_dict={x : image_input})
featuremaps = activation.shape[3]
plt.figure(plt_num, figsize=(15,15))
for featuremap in range(featuremaps):
plt.subplot(6,8, featuremap+1) # sets the number of feature maps to show on each row and column
plt.title('FeatureMap ' + str(featuremap)) # displays the feature map number
if activation_min != -1 & activation_max != -1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmin =activation_min, vmax=activation_max, cmap="gray")
elif activation_max != -1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmax=activation_max, cmap="gray")
elif activation_min !=-1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmin=activation_min, cmap="gray")
else:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", cmap="gray")